In July 2024 I wrote a five-part series on data center energy use. Two of those posts focused on the thorny question of how data centers would contract for power in settings where utility regulation and timelines were too slow to enable the data center owner to achieve their desired speed-to-power, and how transaction cost economics and the literature on the make-or-buy decision helps us understand why those contractual forms were likely to emerge. A year later, let’s take stock and get an update.

The Transaction Cost Lens: Why Make-or-Buy Still Matters

Transaction-cost economics (TCE) has long explained why firms choose either to produce inputs internally (make) or purchase them externally (buy). According to Oliver Williamson’s foundational analysis, firms adopt governance structures that minimize the sum of production and transaction costs. Key factors like asset specificity, uncertainty, and frequency shape this decision-making process.

Asset specificity is the degree to which an investment is tailored to a particular transaction or partner, limiting its value outside of that specific context or relationship. The canonical example of asset specificity was the contract between General Motors and Fisher Body, in which Fisher had to invest in GM-specific metal stamping machines that would be useless to them if GM cancelled the contract, but that they could also use to hold up negotiations (a case discussed in the seminal Klein, Crawford, Alchian (1978) article and debated by TCE scholars since). Highly specific assets, such as specialized equipment, infrastructure, or site-specific investments, create vulnerability to ex post opportunistic behavior by the contract partner, thus encouraging firms to integrate vertically or employ complex contracts to protect against exploitation. The frequency or repeated nature of transactions also influences governance choices; frequent transactions foster trust and relational contracting, reducing transaction costs and increasing the feasibility of hybrid governance structures.

Data centers exemplify the power of this TCE framework. They represent billions of dollars in immovable, highly specialized capital anchored to specific geographic locations. Coupled with the current electricity demand uncertainty driven by AI development, these factors create conditions ripe for innovative governance forms situated between pure market transactions and full vertical integration. Make-or-buy is a continuum, not a binary.

What a Difference a Year Makes

Over the past year, major hyperscale data centers have evolved their electricity procurement strategies, reshaping relationships with both utilities and third-party providers in novel ways and in different states, depending on the regulatory structure in each state.

Three converging forces have reshaped the economics of data‑center power procurement. First, explosive load growth has moved electricity from a fungible commodity to a mission‑critical production factor. Accenture now projects that U.S. data‑center demand will increase from 176 TWh in 2023 to 413–509 TWh by 2030, with generative‑AI workloads alone responsible for roughly 40 percent of that increase, multiplying the frequency and value of supply transactions. Second, interconnection bottlenecks have intensified holdup risk; in a series of show‑cause orders issued between January and April 2025 (Dockets EL25‑49, AD24‑11, EL25‑20), FERC concluded that PJM’s tariff “lacks clarity” for large‑load co‑location and effectively stalled more than 3 GW of new demand in the queue (Federal Energy Regulatory Commission). Third, mounting policy pressure for reliability and decarbonization has raised the shadow cost of relying solely on utility supply, both due to timing and affordability considerations among regulators and consumer advocates. For example, Berkeley Lab’s Natalie Mims Frick’s presentation at the June 2025 NASUCA conference emphasized the importance of innovative rate designs that insulate other customers (particularly residential customers) from costs associated with large‑load uncertainty. Transaction‑cost economics tells us that rising asset specificity and uncertainty in this environment will push data center owners away from market purchases and toward hybrids or vertical integration to safeguard quasi‑rents and speed adaptation.

Monetizing Existing Nuclear Capacity

Meta’s recent 20-year power purchase agreement (PPA) with Constellation for output from Illinois’ Clinton nuclear plant represents an important innovation. Rather than waiting years for interconnection studies or new plant builds, Meta secured access to 24/7 carbon-free power by monetizing additional capacity created through plant upgrades. Here, asset specificity (existing nuclear generation infrastructure) aligns well with Meta’s strategic need for stable, low-carbon electricity in a setting in which they can be confident that their demand for electricity will justify these purchases over 20 years.

Joint Ventures as Strategic Flexibility

Google’s joint venture with Elementl Power to develop 1.8 GW of advanced nuclear across three sites demonstrates another facet of the make-or-buy decision. Amid technological uncertainty and regulatory ambiguity surrounding advanced reactors, Google chose partial integration through joint ownership. This governance form provides flexibility, capturing the option value inherent in evolving nuclear technology without committing to outright ownership.

Adapting to Regulatory Realities

Amazon Web Services’ (AWS) revised deal with Talen Energy provides an example of how regulatory factors shape transactional structures. Initially designed as a behind-the-meter project, FERC regulatory hurdles forced AWS to restructure the agreementinto a front-of-meter retail supply deal using local utility infrastructure. By adapting to federal jurisdictional boundaries, AWS reduced regulatory uncertainty, accelerating project timelines and mitigating risk.

Co-development to Manage Risk

Amazon’s memorandum of understanding (MOU) with Dominion Energy for small modular reactors (SMRs) in Virginia represents yet another hybrid approach. This intentionally incomplete contract allocates development and licensing risks, maintaining future flexibility without prematurely locking into a specific technology or capital structure.

Bridging the Gap with Microgrids

Microsoft’s partnership with Enchanted Rock in San Jose illustrates the growing role of microgrids as an intermediate solution. Deploying RNG-fueled microgrids provides Microsoft immediate reliability and autonomy, circumventing lengthy utility upgrade timelines. Even after eventual grid integration, the microgrid retains value through continued islanding capabilities, underscoring a key TCE insight: hybrid solutions effectively manage both asset specificity and time-sensitive uncertainty.

Emergence of a New Contractual Menu

These diverse transactions were catalyzed by ongoing PJM-FERC “co-location” proceedings, effectively creating a menu of governance options classified as firm, generator-contingent, and interruptible services (as articulated in the FERC filing from Constellation). The contractual diversity that has resulted mirrors Williamson’s market-to-hierarchy (buy-to-make) continuum:

• Pure Buy (standard tariffs): In restructured states, hyperscalers typically buy from independent power producers (IPPs), while the local utility remains responsible for grid interconnection. This process can often delay projects significantly, misaligned with hyperscalers’ rapid speed-to-power timelines. In vertically-integrated states, utilities manage both generation and interconnection, which similarly can introduce substantial delays.
• Hybrid Contracts (generator-contingent): Seen in Meta and Amazon’s nuclear PPAs and MOUs, these contracts mitigate specificity and uncertainty risks through relational arrangements and longer-term contracts, as an alternative to either pure buy or self generation. These hybrids can be diverse and can be customized to the parties’ preferences and to the specific regulatory context in which that relationship will be situated.
• Self-generation/Interruptible Arrangements: Microsoft’s microgrid strategy typifies this approach, prioritizing rapid service deployment over firm grid connection.

The Economics Driving New Contractual Forms

Why are these innovative governance structures proliferating now? Using TCE and the systematic analysis of the make-or-buy decision suggests the following:

• Asset Specificity: Hyperscale operators anchor their data centers physically and contractually to unique, immovable assets like nuclear plants, safeguarding against price volatility and market hold-ups.
• Uncertainty and Flexibility: Joint ventures (Google–Elementl) provide valuable flexibility amidst technological and regulatory unknowns.
• Speed-to-Service: Lengthy grid interconnection queues (often 36+ months) push firms toward microgrids or contingent contracts as immediate solutions.
• Regulatory Risk Management: Firms strategically restructure deals, as seen in AWS’s adaptation at Susquehanna, to navigate jurisdictional hurdles efficiently.
• Control Without Capital Intensity: PPAs and joint ventures grant critical residual control rights to hyperscalers, mitigating transactional friction without incurring full ownership costs.

Policy Innovations and Utility Responses

Regulators, utilities, and consumer advocates are now working with data center owners to develop innovative tariffs to align with these emerging contractual forms. Tariffs designed for large loads increasingly incorporate minimum load factors, backup-power obligations, and clean-energy requirements, recognizing data centers’ strategic importance and unique cost profiles. These tariffs and the evolving issues of data center flexibility are a topic for another day!

Utilities face a choice: adapt by creating transparent, flexible tariffs that mirror these hybrid governance forms or risk obsolescence as data centers integrate around them. Regulators must navigate between equitable and affordable grid access across all customers and accommodating critical economic infrastructure, framing large-load tariff structures that both recover incremental costs and incentivize investment. Those regulatory objectives are essentially conceptually the same as they were in the 20th century, although with innovation and dynamism the ways of achieving those objectives can and should change.

Beyond Transaction Costs

The past year’s contractual innovations underscore how hyperscale data centers are reshaping power systems in fundamental ways through strategic applications of TCE principles. Far from simply choosing between market or hierarchy, these actors are redesigning the market itself, creating new hybrid forms tailored to their economic and operational realities.

This ongoing transformation is the real story, driven by asset specificity, uncertainty, regulatory context, and timing, with profound implications for the future of electricity procurement, grid management, and regulatory policy.

This piece was originally published on Lynne’s Substack, Knowledge Problem. If you enjoyed the piece, please consider subscribing here.